The Universe of Discourse

Thu, 02 Nov 2017

A few years back I wrote a couple of articles about the extremely poor macro plugin I wrote for Blosxom. ([1] [2]). The feature-poorness of the macro system is itself the system's principal feature, since it gives the system simple behavior and simple implementation. Sometimes this poverty means I have to use odd workarounds to get it to do what I want, but they are always simple workarounds and it is never hard to figure out why it didn't do what I wanted.

Yesterday, though, I got stuck. I had defined a macro, ->, which the macro system would replace with →. Fine. But later in the file I had an HTML comment:

    <!-- blah blah -->


and the macro plugin duly transformed this to

    <!-- blah blah -→


creating an unterminated comment.

Normally the way I would deal with this would be to change the name of the macro from -> to something else that did not conflict with HTML syntax, but I was curiously resistant to this. I could not think of anything I wanted to use instead. (After a full night's sleep, it seems to me that => would have been just fine.)

I then tried replacing --> with &#45;&#45;&#62;. I didn't expect this to work, and indeed it didn't. Those &# escapes only work for text parts of an HTML document, and cannot be used for HTML syntax. They remove the special meaning of character sequences, which is the opposite of what I need here.

Eventually, I just deleted the comment and moved on. That worked, although it was obviously suboptimal. I was too tired to think, and I just wanted the problem out of the way. I wish I had been a little less impulsive, because there are at least two other solutions I overlooked:

• I could have defined an END-HTML-COMMENT macro that expanded to --> and then used it at the end of the comment. The results of macro expansion are never re-expanded, so the resulting --> would not have been subject to further replacement. This is ugly, but would have done the job, and I have used this sort of technique before.

• Despite its smallness, the macro plugin has had feature creep over the years, and some of these features I put in and then forgot about. (Bad programmer! No cookie!) One of these lost features is the directive #undefall, which instructs the plugin to forget all the macro definitions. I could have solved the problem by sticking this in just before the broken comment.

(The plugin does not (yet) have a selective #undef. It would be easy to add, but the module has too many features already.)

Yesterday morning I asked my co-workers if there was an alternative HTML comment syntax, or some way to modify the comment ending sequence so that the macro plugin wouldn't spoil it. (I think there isn't, and a short look at the HTML 5.0 standard didn't suggest any workaround.)

One of the co-workers was Tye McQueen. He said that as far as he knew there was no fix to the HTML comments that was like what I had asked for. He asked whether I could define a second macro, -->, which would expand to -->.

I carefully explained why this wouldn't work: when two macro definitions share a prefix, and they both match, the macro system does not make any guarantee about which substitution it will perform. If there are two overlapping macros, say:

    #define pert curt
#define per  bloods


then the string pertains might turn into curtains (if the first substitution is performed) or into bloodstains (if the second is). But you don't know which it will be, and it might be different each time. I could fix this, but at present I prefer the answer to be “don't define a macro that is a prefix of another macro.”

Then M. McQueen gently pointed out that -> is not a prefix of -->. It is a suffix. My objection does not apply, and his suggestion solves the problem.

When I write an Oops post I try to think about what lesson I can learn from the mistake. This time there isn't too much, but I do have a couple of ideas:

1. Before giving up on some piece of software I've written, look to see if I foresaw the problem and put in a feature to deal with it. I have a notion that this would work surprisingly often.

There are two ways to deal with the problem of writing software with too many features, and I have worked on this problem for many years from only one direction: try to write fewer features. But I could also come at it from the opposite direction: try to make better use of the features that I do write.

It feels odd that this should seem to me like a novel idea, but there it is.

2. Try to get more sleep.

(I haven't published an oops article in far too long, and it certainly isn't because I haven't been making mistakes. I will try to keep it in mind.)